Examples provide a multi-stage cluster component that performs a multi-stage clustering analysis on a plurality of items in a category associated with a selected item using a set of interrelationship factors. The multi-stage cluster component generates a cluster of non-substitute item-pairs, a cluster of traditional substitute item-pairs, and a cluster of variety item-pairs. The set of interrelationship factors includes at least one of measure of association, brand similarity, pack-size similarity, demographic similarity, item description similarity, lift, and/or percentage same-basket variable. A propensity score is generated for each item-pair. The propensity score is utilized to identify traditional substitute items and variety substitute items. Each substitute item is ranked based on the generated propensity score. The ranking is used to identify potential low-performance items for removal from inventory.